5 questions to ask before hiring any AI project

Every week, the average company receives at least one AI project proposal. The material arrives polished, packed with technical jargon and promises of transformation. The owner sits through the presentation, asks one or two questions about price and timeline, and makes the decision with the wrong information.

The problem isn't the technology. The problem is that the owner doesn't know what to ask. And the vendor will rarely fill that gap on their own.

Below are the five questions that actually make a difference. Not because they're difficult, but because the answers reveal whether the project will work in your operation or become an expensive tool that nobody uses.

"What specific task does this solve, and how much time does my team currently waste on it?"

Every AI project needs to start with a concrete, measurable pain point. If the vendor can't name the task, the proposal is too generic to work.

Think of it this way: if your team spends twenty minutes sorting customer emails before responding, two people, ten times a day, that's real time that comes back to the team once the problem is solved. The vendor has to go into that level of detail. If they only talk about "increasing efficiency" or "improving the customer experience" without naming the task, ask them to walk through a specific example. If they can't, the project probably wasn't designed for your company.

What the answer reveals: whether the vendor understands your business or is simply recycling a standard proposal.

"What data will the solution work with, and does that data already exist in my company?"

AI learns from information. If the information doesn't exist, is disorganized, or is locked across three different systems that don't talk to each other, the project stalls before it even begins.

In most conversations we have with business owners, the data exists, but it's scattered. A spreadsheet here, a management system there, a history buried in the inbox of a salesperson who left six months ago. Before hiring anything, you need to know whether the vendor has already mapped out where the data will come from, and who inside your company will be responsible for keeping that data clean and up to date.

If the answer is "we'll use whatever data you have," ask what happens when the data is wrong. The answer to that second question says a lot about how committed the vendor actually is to delivering results.

Red flag: a vendor who doesn't ask about your data before presenting the price.

"What happens when the solution makes a mistake?"

Every AI solution makes mistakes. That's not a flaw, it's a feature. The real question is what happens next.

A customer service assistant on WhatsApp will, at some point, provide incorrect information. A system that suggests stock items for replenishment will, occasionally, recommend the wrong quantity. The project needs to account for this. There has to be someone on your team reviewing critical cases, a clear path to correcting the error before it reaches the customer, and a way for the solution to learn from what it got wrong.

Ask the vendor: "Give me an example of a mistake this solution could make and walk me through what happens next." An honest answer includes both the error and the correction process. The answer you don't want to hear is "the system is very accurate, errors are rare."

What the answer reveals: whether the vendor is selling the tool or selling the outcome.

"Who operates this after the project is delivered?"

This is the question most owners forget to ask, and it's the one that most determines whether the investment will last.

Every project has a delivery moment. What happens the following week? If the solution requires someone on your team to feed it, monitor it, and adjust it, that person needs to exist, have the time, and know what they're doing. If maintenance stays on the vendor's side, what contract guarantees that? For how long? At what cost?

In practice, we frequently see two scenarios. In the first, the project is delivered, the vendor leaves, and the solution slowly degrades because no one at the company knows how to work with it. In the second, a maintenance contract exists, but the scope is vague, and every adjustment turns into a new charge.

Before signing, ask to see the support contract and have the vendor explain in plain language what is and isn't included.

What the answer reveals: whether the vendor thought about your business beyond the initial delivery.

"How will I know it worked?"

A project without a success criterion defined upfront is a project without accountability. You have no way of knowing whether the result was worth the investment, and the vendor has no way of being held responsible.

The criterion needs to be simple and measurable by your own team, without depending on the vendor to calculate it. "Average customer response time dropped from four hours to forty minutes." "Rework on invoice review hit zero." "The customer service team stopped working overtime on Fridays." Those are success criteria. "Improved user experience" is not.

If the vendor resists defining that criterion before the project begins, the problem is one of alignment, not technology.

Red flag: a vendor who proposes measuring success with metrics that only they can access.

What to do with these answers

Bring the five questions to your next meeting with any technology vendor. Not as a test, but as a conversation. A good vendor will be happy to answer them, because they've already mapped all of this out before walking in the door.

Evasive answers, on the other hand, are information in themselves. They signal that the project was designed for the vendor's portfolio, not for your operation.

The decision to invest in AI doesn't have to be risky. It becomes risky when it's made using information from someone who wants to sell, not from someone who will have to operate it.